Comment 7 for bug 1467927

Revision history for this message
OpenStack Infra (hudson-openstack) wrote : Fix merged to nova (master)

Reviewed: https://review.openstack.org/229575
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=d5bed8fad9d51dd0d4ba9cf452c6f5c8c543b264
Submitter: Jenkins
Branch: master

commit d5bed8fad9d51dd0d4ba9cf452c6f5c8c543b264
Author: Nikola Dipanov <email address hidden>
Date: Wed Sep 30 19:30:59 2015 +0100

    Fix CPU pinning for odd number of CPUs w hyperthreading

    Previous logic was not handling the cases when the instance cell has an
    odd number of vcpus that is larger than any sibling set on the host. The
    logic would (accidentally) either reject to pin such an instance even
    though there is ample free cores, or would (in case there were enough
    free sibling sets on the host) spread the instance instead of pack it
    onto siblings as the default policy suggests.

    This patch fixes some incorrect assumptions in the code, while also
    simplifying it. As an added bonus - we still attempt to expose
    (via the Topology, and this time correctly) the largest possible number
    of threads that we can expose to the instance.

    Finally - we add some more comments to clear up the intent behind the
    current packing logic with pointers how it could be tweaked to achieve
    different results in the future.

    Change-Id: I2c0b3b250ffb1a7483299df13b317cdb24f8141d
    Co-Authored-By: Stephen Finucane <email address hidden>
    Closes-bug: 1501358
    Closes-bug: 1467927